A non-intrusive method to add finite element-based random variables to a probabilistic design code

Abstract

Probabilistic design computer programs are usually tailored to a specific mechanics problem and do not involve general-purpose finite element analysis due to efficiency and simplicity concerns. However, as these design codes become more widespread, the capability to extend the applicability of the design code by considering additional finite element-based random variables become more desirable and important. This research describes a methodology such that additional finite element-based random variables can be considered without modifying the design code software, that is, the methodology is “non-intrusive” and existing software can be extended without access to the source code. Probabilistic sensitivity measures have been developed for both the original and additional random variables such that the significance of the random variables can be determined. Efficiency issues are addressed through the construction and subsequent verification using response surface methods. The methodology is general in that it can be applied to any probabilistic design code and any finite element model as long as the original and additional random variables are independent—there can be correlation within a group. The methodology is demonstrated using a probabilistic fatigue application whereby a probabilistic design code is integrated in an automated fashion with a commercial finite element program.